sim.loss issue : "ValidationError: input data: should have rank 3 (batch_size, n_steps, dimensions), found rank 4" about nengo-dl HOT 5 CLOSED

Rank refers to the number of dimensions in the array. So, e.g., an array with shape (5,) is rank 1, (5, 3) would be rank 2, etc. So that error is telling you that you have an extra dimension in your input data. In that line print(x_train.shape..., I'd guess that that is printing out a shape with 4 elements (likely because the two-dimensional images have not been flattened).

from nengo-dl.

Ah, yes. Attached are some tests I ran after loading both datasets (prior to adding the time dimension). They seem so similar (screenshot), so was not sure why the mnist fashion would throw an extra dimension error when running sim. After poring through data very thoroughly, was able to sort it out- issue was with each row in mnist fashion image being stored as an array, rather than one large array.

After restructuring data, got spiking working with fashion-mnist! I appreciate the help!

from nengo-dl.

That is what I had assumed, but when I print("x_train shape:", x_train.shape, "y_train shape:", y_train.shape) , it returns x_train shape: (60000, 28, 28) y_train shape: (60000,) So it appears that the x_train is 3 dim. I appreciate your help, excited to start working with nengo-dl more.

from nengo-dl.

Oh also keep in mind that if you're following the spiking-mnist example then later on you're adding an extra dimension (for time), so your shape would become (60000, 1, 28, 28).

from nengo-dl.

Hi @joh10891!

I am also facing the same issue, so can you please share the code of how you restructured the data that helped you resolving the issue ?

from nengo-dl.